Dengue is a positive-sense RNA virus belonging to the Flaviviridae family. There are four distinct serotypes — dengue virus serotype 1 (DENV1) to DENV4 — and several genotypes within each serotype. Since the 1950s, there has been considerable expansion in the geographic spread of dengue and an exponential rise in disease incidence. It is estimated that the annual global incidence is 390 million cases, of which 96 million are clinically apparent1. There are several factors that drive this pandemic, including globalization, the spread of the Aedes mosquito vector, inadequately planned urbanization and the absence of a licensed vaccine or anti-dengue therapeutics2.

The clinical phenotype of dengue can vary depending on several factors, of which age, genetic predisposition and background immunity are major determinants. However, most clinical infections result in a self- limiting febrile illness termed dengue fever. The hallmark feature of severe disease is increased capillary permeability, causing plasma leakage, which can lead to haemodynamic compromise and dengue shock syndrome (DSS) (Box 1). If untreated, severe disease can lead to a mortality rate of up to 20% of affected individuals but with expert management and primarily careful fluid replacement, this can be reduced to less than 1%2.

The pathogenesis of severe disease is thought to be due, at least in part, to immune mechanisms in which antibody enhancement and T cell immunopathology probably have key roles. These theories have served to explain the observation that the development of severe disease is associated with serial infection with dengue viruses of different serotypes. However, because of this risk of severe disease in sequential infections, developing a safe vaccine that provides protection against all four serotypes has been challenging. There have been several advances in understanding and tackling dengue virus infection over the past decade, including viral epitope mapping and defining conserved neutralizing antibodies, as well as the publication of the first Phase III vaccine efficacy trials. This Review focuses on the adaptive immune response to dengue virus infection, the challenges of developing a dengue virus vaccine and new insights from the study of human monoclonal antibodies (mAbs).

Dengue virus

Dengue virus is a relatively simple positive-sense single-stranded RNA virus that is 50 nm in diameter and has three structural proteins — capsid (C) protein, precursor membrane (prM) protein and envelope (E) protein — as well as seven non-structural (NS) proteins termed NS1, NS2A, NS2B, NS3, NS4A, NS4B and NS5 (Fig. 1). Studies using X-ray crystallography and cryo-electron microscopy (cryo-EM) have shed considerable light on structural aspects of the flaviviral life cycle3,4,5,6,7,8,9 (discussed later).

Figure 1: The life cycle of dengue virus.
figure 1

Dengue virus exists as a number of different viral forms depending on the degree of precursor membrane (prM) protein cleavage. Fully immature dengue virus particles contain a full complement of prM proteins and are non-infectious, whereas all prM proteins are cleaved in fully mature virus particles. A number of intermediate, partially mature forms exist in which some prM proteins have been cleaved and some remain intact. Fully mature and some of the partially mature virus particles are infectious (step 1). The dengue viral replication process begins when a virion attaches directly to a diverse group of host cell receptors or when the Fc portion of a dengue virus-containing immune complex attaches to a Fc receptor on the target cells (step 2) and subsequently enters the cell by receptor-mediated endocytosis (step 3). Acidification of the endosomal vesicles triggers conformational changes in the virion, resulting in an irreversible trimerization of the viral envelope (E) protein (not shown). This exposes the fusion peptide and mediates fusion between the viral and the endosomal membranes, allowing the release of the nucleocapsid into the cytoplasm. The viral RNA is released into the cytoplasm and presented to the rough endoplasmic reticulum (ER) (step 4). At the ER, viral RNA is translated into a single polyprotein that is processed by viral and host proteases (step 5). After the viral replication complex is synthesized, viral RNA translation switches off, and RNA synthesis begins by the transcription of an antisense viral RNA followed by the amplification of viral RNA (step 6). The newly synthesized RNA is subsequently packaged by capsid (C) protein, forming a nucleocapsid (step 7). Virus assembly occurs on the surface of the ER when the nucleocapsid buds into the ER lumen, resulting in non-infectious, immature viral particles (step 8). Immature viral particles are transported through the Golgi into the trans-Golgi network, where acidification induces conformational changes of the virion and exposes the furin cleavage sites. The host protease furin cleaves between pr protein and M protein, with the pr protein remaining associated until the virion is released in the neutral pH of the extracellular millieu (step 9). DC-SIGN, dendritic cell-specific ICAM3-grabbing non-integrin; NS proteins, non-structural proteins.

PowerPoint slide

E protein and prM protein form the glycoprotein shell of the virus, and E protein is responsible for host cell binding and entry10. Dengue E protein has two N-linked glycosylation sites and is divided into three domains termed EDI, EDII and EDIII5. During viral assembly in the endoplasmic reticulum, 180 copies of E protein associate with 180 copies of prM protein to form 60 trimeric (heterohexameric) spikes, which gives immature virions their characteristic spiky appearance10 (Fig. 2a). In the trans-Golgi network, prM protein is cleaved at a membrane proximal site by host-encoded furin protease, resulting in a membrane-anchored M stump and a pr molecule that remains associated with the virion until it is secreted. Upon secretion from the host cell, pr protein disassociates to leave the 'mature virion', which has a smooth structure (Fig. 2b).

Figure 2: Structure of the dengue viral particle.
figure 2

a | The structure of an immature dengue viral particle shows the arrangement of envelope (E) proteins in pink and precursor membrane (prM) proteins in light blue. b | The mature dengue virus shows the arrangement of the E proteins into a smooth particle. Figure in part a from Ref. 149, Nature Publishing Group. Figure in part b reprinted with permission from Ref. 3, Elsevier.

PowerPoint slide

The cleavage of prM protein is not complete in all dengue virions, leaving a proportion of partially mature dengue virions that contain a varying amount of cleaved and uncleaved prM protein11,12,13,14. The cleavage of prM protein is more efficient in certain cell types, particularly primary human cells such as dendritic cells, compared with the cleavage event in insect cells or tumour cell lines such as Vero15,16. Fully immature dengue viruses contain regular trimeric E–prM protein spikes and are non-infectious17,18. By contrast, the partially mature forms, some of which are infectious, have a less regular structure, with areas that are spiky and contain prM protein–E protein trimers and areas that are smooth and are proposed to contain E protein dimers19.

Finally, virions attach to cells, possibly through EDIII10. Several host proteins — such as heparan sulfate, dendritic cell-specific ICAM3-grabbing non-integrin (DC-SIGN; also known as CD209), heat shock protein 70 kDa (HSP70), HSP90, mannose receptor, CD14, T cell immunoglobulin and mucin domains (TIMs; also known as HAVCRs), TAM receptor protein tyrosine kinases and laminin — have been suggested to have a role in dengue virus binding, but there is no single receptor yet defined that is necessary for entry20. Following endocytosis, acidification in the early endosome triggers a dramatic reorganization of the viral envelope, from the dimer form to a new trimeric conformation. This exposes the fusion loop at the tips of the trimeric spikes, which then fuses with the endocytic membrane and allows the viral nucleic acid to be released into the host cell cytoplasm10 (Fig. 1).

Immune involvement of dengue pathogenesis

There are several key observations that suggest a role for the immune system in dengue pathogenesis: first, severe dengue usually, but not exclusively, occurs with secondary infection, implying that priming of the adaptive immune system to one dengue virus serotype leads to disease enhancement during a secondary infection with a different serotype2,21,22. Second, the peak in symptoms occurs when viral loads are rapidly decreasing, implying that immunopathology may be driven as a consequence of viral control23,24. Third, higher viral loads and/or soluble NS1 levels in patients with dengue haemorrhagic fever versus patients with dengue fever predispose to increased disease severity, which manifests when the virus is controlled and viral loads fall23,24,25. Finally, the peak in symptoms, vascular leak and viral control coincide with a cytokine storm, which is characterized by high circulating levels of many pro-inflammatory mediators such as tumour necrosis factor (TNF), soluble TNF receptor 1 (sTNFR1), sTNFR2, interferon-γ (IFNγ), CXC-chemokine ligand 8 (CXCL8), CXCL9, CXCL10, CXCL11, CC-chemokine ligand 5 (CCL5) and vascular endothelial growth factor A (VEGFA), as well as the anti-inflammatory cytokine interleukin-10 (IL-10)24,26,27,28,29,30. This cytokine storm has been proposed to trigger increased vascular permeability and resolves relatively quickly upon patient convalescence.

The exact mechanisms underlying the capillary leak are probably more complex than a cytokine storm, which is also associated with other diseases with very different clinical phenotypes, such as graft-versus-host disease31, septic shock32 and influenza virus infection33. Dengue virus may cause increased vascular permeability through binding to the endothelial glycocalyx layer, which lines the luminal surface of microvessels, providing vital barrier functions to capillaries. Both the virus and NS1 protein have been shown to bind to heparan sulfate, a major component of the glycocalyx layer34,35, which may in turn alter the permeability of capillaries (Fig. 3).

Figure 3: Adaptive immune responses to dengue virus infection.
figure 3

a | The phenomenon of original antigenic sin occurs when memory T cell activation in a second infection (for example, with dengue virus serotype 2 (DENV2) in middle and right panels) leads to the domination of that response by cross-reactive T cells generated during a primary infection (for example, DENV1 in the left panel) with a different dengue virus serotype. Three types of cross-reactive T cells are illustrated: type I produce only inflammatory cytokines during secondary infection, type II can degranulate and potentially kill DENV2-infected target cells effectively, and type III cross-reactive T cells are of low avidity and can recognize DENV2-infected cells but do not induce a response. If there is a high proportion of type I cross-reactive T cells that are skewed to inflammatory cytokine production without cytotoxicity, it may predispose to immunopathology on secondary infection (middle panel), whereas if there is a predominance of type II cells that can degranulate (right panel), the virus may be more rapidly cleared. b | The mixture of newly produced dengue virions can be neutralized by the optimum concentration of dengue virus-specific antibodies. In addition, dengue virus proteins — such as envelope (E) protein, precursor membrane (prM) protein and non-structural protein 1 (NS1) — presented on the surface of dengue virus-infected cells, or soluble NS1, can be recognized by dengue virus-specific antibodies. These antibodies can promote complement-dependent cell lysis following complement activation by the classical pathway. Soluble NS1 can also bind to heparan sulfate, a major component of the endothelial glycocalyx layer that regulates capillary permeability. The subsequent recognition by NS1-specific antibodies triggers complement activation and anaphylatoxin formation, which may contribute to the disruption of this layer and vascular leak. Presentation of viral peptides on MHC class I or MHC class II molecules activates DENV-specific CD8+ T cells or CD4+ T cells, respectively. The recognition of MHC–peptide complexes by DENV-specific T cells or effective type II cross-reactive T cells (only cross-reactive CD8+ T cells are shown for simplicity) results in target cell lysis with or without cytokine production, whereas peptide recognition by ineffective DENV cross-reactive T cells (type I and type III) causes only cytokine production or no response. Sub-neutralizing concentrations of E- and/or prM-specific antibodies can bind to the dengue virus and mediate dengue virus infection of Fc receptor-expressing target cells, such as monocytes and macrophages, resulting in antibody-dependent enhancement (ADE) and increased viral load. The cytokines secreted by dengue virus-infected cells and those produced by T cells can be amplified by the high viral load and cause a cytokine storm and plasma leakage in patients infected with dengue virus. c | prM-specific antibodies show poor neutralization but are potent at inducing ADE. Fully mature particles contain no prM protein and will not be neutralized by prM-specific antibodies, whereas low-prM protein-containing particles do not contain enough prM protein for neutralization but can be opsonized for ADE. Some partially mature particles with higher levels of prM protein will be neutralized, whereas fully immature particles are non-infectious but can be made to infect cells via ADE. ER, endoplasmic reticulum; TCR, T cell receptor.

PowerPoint slide

NS1-mediated complement activation may also play a part in the altered capillary permeability through the generation of the anaphylatoxins C3a and C5a and the terminal complement complex C5b–C9 (Ref. 23). High C5b–C9 levels in plasma and pleural fluid have been demonstrated in patients with dengue shock, and patients who were high producers of C5b–C9 had more signs of plasma leakage23. Furthermore, NS1 may have an immunomodulating role by binding to and reducing the functional capacity of C4, thereby altering the classical and lectin complement pathways36.

Innate immune cells also have a role in dengue pathogenesis; for example, several studies suggest that mast cells have a role in both protection against and pathogenesis of dengue. Upon interaction with dengue virus, mast cells release chemokines that recruit T cells, natural killer (NK) cells and NKT cells to control the infection37. By contrast, the interaction also leads to the secretion of mediators such as chymases — proteases that may contribute to vascular leakage. The level of chymases is correlated with disease severity37.

Overall, the well-established association of severe dengue with secondary infection, the evidence of extreme immune activation in severe infection and the coincidence of viral control with the peak of symptoms has led to the detailed study of both T cell- and B cell-mediated adaptive immune responses, which are reviewed here.

The role of T cells in dengue pathogenesis. The association of severe dengue symptoms with a rapid decline in viral loads and a peak of pro-inflammatory cytokine secretion have led to the proposal of a role for a T cell-mediated immune response in driving immunopathology in severe dengue. There has been much study of T cell responses in dengue, but the contribution of these responses to pathogenesis and protection remains controversial38. Reports have demonstrated potent T cell responses in dengue-infected individuals, and many dengue T cell epitopes have been mapped in humans and mouse models39,40,41,42,43,44,45,46. Responses to NS3 seem to be immunodominant in humans, although CD8+ and CD4+ T cells that respond to epitopes across the whole virus proteome have been observed39,41,44,46,47,48,49,50.

The magnitude of the T cell response positively correlates with disease severity39,42,43,51,52 with many cells during the acute disease displaying markers of both proliferation and programmed cell death42. Some scientists have questioned the relevance of T cells to dengue pathogenesis and control because of the relatively low numbers of circulating T cells seen during acute infection that increase rapidly following viral clearance42,53,54. However, a recent study showed that skin blister fluid taken from dengue-infected patients contained more dengue-specific T cells than peripheral blood, suggesting that dengue-specific T cells may migrate from the peripheral blood to the skin during acute infection and return upon viral clearance55.

During secondary dengue virus infection, responding T cell populations rapidly expand and can account for up to 20% of circulating T cells53, and cross-reactive T cells in dengue have been well studied39,42,43,46,50,54,56,57. Using double tetramer staining with tetramers containing peptides from either the primary or secondary infecting viruses, we and others have studied the cross-reactivity of CD8+ T cell responses42,43,54. These studies provide evidence for original antigenic sin in secondary T cell responses to dengue virus. During a secondary infection, there is a large increase in the number of T cells that bind to tetramers loaded with peptides derived from the primary dengue virus infection42. Some of these cells cross-react with tetramers loaded with peptide derived from the secondary infecting virus, whereas others show lower avidity for the tetramer from the secondary virus and no functional responses to the secondary virus-derived peptides. Original antigenic sin has previously been reported for CD8+ T cell responses to influenza virus58 and lymphocytic choriomeningitis virus (LCMV)59, and also appears to operate in dengue virus infection42. Some dengue-specific T cell populations that expand during secondary infection have been shown to be fully cross-reactive to antigens of the serotypes of the primary and secondary infections, whereas other populations show preference for the primary-infecting serotype42,43,50. The expansion of these low-avidity cells specific for the primary infection may delay viral clearance and thereby lead to higher viral loads.

In another study, the presence of original antigenic sin was demonstrated in the dengue T cell response, but the magnitude of the response did not seem to be compromised44. This study differed from the one described earlier in that it measured functional responses to peptide stimulation and did not measure non-functional cells with tetramer staining or formally show cross-reactivity, which was inferred on the basis of the similarity of peptide sequences. Thus, in the absence of a good animal model of disease, it remains controversial whether the expansion of low-avidity cross-reactive T cells in secondary dengue infection contributes to disease pathogenesis.

Study of the function of dengue-specific T cells has revealed an interesting difference between mild and severe infections. During milder dengue fever illness, CD8+ T cells degranulate but produce lower levels of pro-inflammatory cytokines39. By contrast, during severe primary and secondary infections, a high proportion of CD8+ T cells produce the cytokines TNF and IFNγ but do not undergo marked degranulation, potentially negatively affecting virus control while enhancing disease pathogenesis and immunopathology39. CD4+ T cell populations also expand during dengue virus infection and, similarly to CD8+ T cells, the magnitude of the CD4+ T cell response correlates with disease severity39,46. Interestingly, several studies have reported the generation of dengue-specific CD4+ T cells that express degranulation markers or can lyse peptide-pulsed target cells, which could have a role in the disease39,46,57,60.

As mentioned above, the role of T cells in dengue disease pathogenesis and protection is debated. On the one hand, potent T cell responses skewed to cytokine production and not degranulation are associated with, and potentially drive, severe disease42,43. On the other hand, in milder disease, CD4+ or CD8+ T cells may have a beneficial role in controlling virus replication. An analysis of T cell responses in individuals from a large Sri Lankan dengue cohort has suggested a role for T cells in protecting from dengue: higher-amplitude T cell responses were observed for epitopes presented by protective HLA alleles44. However, most HLA association studies in dengue have been small and not repeated61. Furthermore, depletion and adoptive transfer of CD8+ and CD4+ T cells in mouse models have suggested a role for these cells in reducing viral loads and protecting from severe disease45,62,63. Other investigators who question the importance of T cells in dengue pathogenesis point out that severe dengue can occur during a primary dengue infection in which cross-reactive T cells and original antigenic sin would not be operative64.

Finally, it has been speculated that the suboptimal efficacy of the recent Sanofi Pasteur vaccine in clinical trial (see later) may in part be driven by the lack of a protective T cell response, as all of the NS proteins in this tetravalent vaccine derive from the yellow fever virus. We suggest that the generation of an early T cell response to dengue virus may be protective, whereas the late generation of T cell populations that have a proportion of low-avidity T cells and are skewed to inflammatory cytokine production in the absence of degranulation may predispose to immunopathology in the presence of high viral or antigen loads and contribute to the cytokine storm and vascular leak39 (Fig. 3).

Antibody-dependent enhancement. The theory of antibody-dependent enhancement (ADE) attempts to explain why severe dengue is usually associated with secondary infections in children and adults21,22,65. The ADE hypothesis posits that pre-existing heterologous antibodies generated in response to a primary infection may not be of sufficient avidity or concentration to neutralize a secondary infection by a different dengue serotype, in which the sequence of E protein may vary by 30–35% at the amino acid level. Instead, the virus may be opsonized and targeted for uptake into Fc receptor (FcR)-bearing cells, such as monocytes and macrophages, which are major sites of dengue virus replication in vivo (Fig. 3b). It has also been shown that dengue viruses interact with leukocyte immunoglobulin-like receptor B1, which leads to inhibition of FcR-mediated signalling induced by dengue virus-specific antibody complexes and the downstream induction of antiviral type I IFN-stimulated genes66.

ADE can be readily demonstrated in vitro using FcR-expressing cells and has been shown to drive higher viral loads in both mouse and primate models of dengue virus infection67. ADE has also been suggested to lead to increased disease severity in infants experiencing primary dengue virus infection during the first year of life when the levels of passively transferred maternal dengue-specific antibodies fall below neutralizing levels21,68,69,70. In mouse models, ADE can increase disease severity, leading to vascular leakage and lethality67,71. Polymorphisms in the gene encoding the low-affinity FcR for IgG (FCGR2A) have been suggested to have a role in ADE; there is an association of the H/H and H/R alleles at position 131 with severe infection, whereas the 131 R/R allele is associated with protection from severe disease72,73,74.

Antibodies directed against prM protein can facilitate FcR-mediated uptake of immature viral particles which, although non-infectious in the absence of antibody, can undergo furin-mediated prM protein cleavage following endocytosis in the host cell, rendering them infectious15,18,75. A similar phenomenon has also been demonstrated following ADE driven by antibodies specific for E protein76. At sub-neutralizing concentrations, almost all antibodies specific for E and prM proteins can enhance viral uptake, with maximum levels of enhancement occurring at roughly half the concentration required for neutralization77,78. In infections with West Nile virus, a structurally related flavivirus, it has been estimated that around 15–29 E protein-specific antibodies bound to a virion will promote ADE78. The exact role of ADE in human infection is much debated and has been a great concern in vaccine studies. The very recent data (described later) from the Sanofi Pasteur vaccine trials may be the first evidence to substantiate this concern.

In summary, antibodies and T cells have been implicated in the generation of both protective and pathogenic responses during dengue virus infection. Optimizing protective responses while minimizing pathogenic responses are the goals for successful vaccines.

Dengue vaccines

The exponential rise in dengue virus infections over the past few decades has made the search for a dengue vaccine an imperative, but achieving this goal has proved enormously challenging. Any successful vaccine would need to induce a protective and durable immune response to all four dengue serotypes, preferably with one or two doses, in individuals who have been unexposed to dengue virus, as well as those who have had a previous dengue virus infection. At the same time, a vaccine would need to avoid eliciting or enhancing pathogenic immune responses.

As primary dengue virus infection does not give long-term protection to re-infection with the other three viral serotypes79,80, it has been generally held that a vaccine will need to induce protective responses against all four serotypes, mandating a tetravalent formulation. Efforts to develop vaccines have been pursued for almost 50 years, beginning in Thailand with work to produce live attenuated viruses by serial passage of viral strains representative of the four serotypes81. A particular challenge has been to develop attenuated forms of the viruses that are not too virulent to induce overt dengue disease but are not too over-attenuated to be unable to incite a protective immune response. Another challenge has been to produce a tetravalent formulation in which all four viruses are delivered together, replicate equally and induce a balanced immune response against all four serotypes rather than resulting in good responses to some serotypes but poor responses to one or more serotypes82,83,84.

There have been substantial advances in recent years in vaccine development with three live attenuated vaccines currently undergoing clinical trials: CYD-tetravalent dengue vaccine (CYD-TDV), National Institutes of Health (NIH) live attenuated tetravalent vaccine (LATV) Δ30 and DENVax85,86,87,88 (Fig. 4). There are also various other vaccines in preclinical studies. The most advanced in clinical trials is the Sanofi Pasteur vaccine CYD-TDV. This is a chimaera vaccine using the related yellow fever 17D vaccine strain as a backbone, with dengue virus prM protein and E protein genes replacing those from yellow fever virus. The vaccine contains a mixture of four recombinant viruses representing each serotype (termed CYD1 to CYD4)89,90. Initial clinical trials demonstrated good serological responses to the vaccine, with seropositivity ranging from 66.5% to 100%91,92,93. However, the overall efficacy (that is, protection from infection) of a Phase IIb trial in Thailand was below expectations, with the lowest efficacy being against DENV2 (Table 1). Phase III trials of this vaccine in Asia and Latin America have been more promising; however, the results still showed suboptimal efficacy ranging from 35% to 78% and, again, the efficacies against DENV2 were lowest94,95. Further analysis revealed that the vaccine gave better protection to vaccinees who were already immune to one or more serotypes before vaccination, and the protection offered by the vaccine against severe and haemorrhagic disease was between 80% and 91%94,95.

Figure 4: Dengue vaccines.
figure 4

a | The Sanofi Pasteur vaccine CYD-TDV contains four chimeric live flaviviruses, each derived from the genome of the yellow fever virus 17D vaccine strain (shown in yellow) with the precursor membrane (prM) and envelope (E) gene segments replaced by the corresponding gene segments of each of the four dengue virus serotypes (DENV1 to DENV4). b | The US National Institutes of Health (NIH) live attenuated tetravalent vaccine (LATV) contains a mixture of four recombinant dengue virus genomes; the DENV2 component is a chimeric dengue virus derived from a DENV4 genome with prM and E gene segments replaced by those of DENV2. The vaccine strains were attenuated by deleting 30 nucleotides (Δ30) from the 3′ untranslated region of the dengue viral genome. c | The DENVax vaccine from Takeda contains a mixture of four recombinant DENV2 genomes, each derived from the genome of an attenuated DENV2 virus with prM and E gene segments replaced by the corresponding gene segments of DENV1, DENV3 and DENV4.

PowerPoint slide

Table 1 Dengue vaccine candidates

The first report from the long-term follow up of the CYD-TDV vaccine trials has recently been published96 and shows that older children in the vaccinated group (aged 9–16 years) continue to benefit. However, in the under-9-year age group, there appears to be an increase in hospitalization at 3 years post-vaccination when compared with control unvaccinated subjects96. This increase may represent disease enhancement, possibly by ADE, in individuals who were uninfected at enrolment and who have been primed but not protected by the dengue vaccine. There is now much urgency in the field to understand why the initially promising in vitro correlates of vaccine immunity85,89,90,91,92,93 did not always translate to good in vivo efficacy94,95,96 and to develop more representative in vitro correlates of protection to inform future trials.

In the Phase IIb trial of CYD-TDV in Thailand, although the efficacy of the vaccine against DENV2 was low, the seropositivity to DENV2 at 1 year after vaccination was high90 (Table 1). Several reasons are postulated for this mismatch: first, the vaccine strain might be a poor immunological match for viral strains circulating during the study period, although preclinical trials showed that the vaccine candidate antigens induced neutralizing antibody activity against a broad range of viral strains85. Second, T cells may be important for protection, and because the CYD-TDV vaccine contains NS elements from yellow fever virus, it may not produce an effective dengue virus-specific T cell response. Third, the CYD-TDV vaccine does not contain dengue virus NS1, which was suggested to induce protective immunity in mouse models97,98. Finally, the current plaque reduction neutralization test (PRNT), which uses viruses generated from cell lines, may not be a good surrogate for protection in vivo, and better, more predictive assays are urgently being sought.

The tetravalent formulation does not mimic the natural situation in which the vast majority of infections are with a single serotype followed by reinfection months to years later79,80. Although studying sequential infections in the population is difficult, surveillance studies suggest symptomatic cases caused by third or fourth infections are rare99,100,101. This implies that following secondary infection there is some degree of cross protection to the remaining serotypes. Inducing a balanced immune response to all four serotypes using a tetravalent formulation is clearly a major challenge, and heterologous prime–boost strategies have been suggested to overcome this. The observation in the recent Sanofi Pasteur trials that efficacy was better in individuals who had previously been exposed to dengue virus, compared with the uninfected cohort, may also be instructive in this regard, implying that the vaccine can boost pre-existing dengue virus immunity but is poor at producing a protective immune response de novo in dengue virus-naive vaccinees94.

Sequential priming and heterologous boosting with the recombinant viruses CYD1 to CYD4 has been investigated in primates102. CYD2 and CYD3 were inherently less immunogenic, with lower numbers of responders, than CYD1 and CYD4. Priming with CYD1 and CYD2 followed by boosting at day 56 with CYD3 and CYD4 gave better responses when compared with priming and boosting with the tetravalent vaccine, especially with regard to responses to DENV2 (Ref. 102). Heterologous priming and boosting has also been investigated in humans, in which CYD1 and CYD3 were used to prime and CYD2 and CYD4 used for boosting at 15 weeks103. In contrast to the primate study, the sequential bivalent immunization strategy was not superior to the repeated tetravalent immunization approach. It remains possible that the timing between priming and boosting and the serotype combinations used in these two studies may have influenced the outcome.

Of the other live attenuated vaccine candidates, a Phase I trial of the NIH LATV Δ30 vaccine, which has a 30-nucleotide deletion in the 3′ untranslated regions of the cloned cDNA (Fig. 4b), showed that seropositivity across all serotypes was over 90% after one dose of the vaccine; however, around 60% of vaccinees developed a rash and 73% experienced viraemia86,104. The tetravalent live attenuated vaccine DENVax (developed by Takeda) has undergone a Phase I trial87, in which more than half of the volunteers developed measurable antibodies to all four virus serotypes after a single dose87,105. Finally, despite the demonstration that the TDEN vaccine candidate developed by the Walter Reed Army Institute of Research (WRAIR) and GlaxoSmithKline induced antibody responses to the four serotypes in a Phase II trial, development has been stopped106.

Despite the difficulties, there remains a pressing need to pursue vaccine candidates as well as to test whether live attenuated vaccines can be improved to provide a more balanced protection, and whether these responses are durable will be determined by longer-term follow up of clinical trials.

New insights from monoclonal antibodies

A large number of mouse antibodies have been generated against dengue virus, and the epitopes for several of these have been mapped by systematic or random mutagenesis to the E protein107,108,109,110,111,112,113,114,115,116. More recently, human mAbs have been generated from dengue virus-infected patients; the main epitopes recognized by mAbs are described in this section (Table 2).

Table 2 Characteristics of human monoclonal antibodies

Antibodies targeting the fusion loop epitope of E protein. The fusion loop epitope (FLE) is a major epitope targeted by antibodies produced in both humans and mice. mAbs specific for the FLE are exemplified by mAb E53, which was originally isolated from West Nile virus-infected mice but is fully cross-reactive with dengue viruses117. Mutation of E protein W101 blocks the binding of antibodies directed to the FLE and reduces the reactivity of polyclonal human serum118,119,120. mAbs specific for FLE show good neutralization of dengue viruses grown in insect cells, which contain high levels of prM protein16. FLE-specific mAbs bind better in the presence of prM protein which, by virtue of changing the virion architecture, better exposes the FLE13. The binding of FLE-specific mAbs to more mature virions produced in primary human cells, which contain lower levels of prM protein, and their subsequent neutralization is not particularly efficient, with neutralization levels typically failing to exceed 60–80% even at high antibody concentrations16. The induction of antibodies specific for FLE, although commonly generated in humans and often cross-reactive between serotypes, may not therefore be the ideal response to target by vaccines. Another antibody, 1C19, which recognizes the BC loop (amino acids 73, 78 and 79) close to the FLE, potently neutralized all four serotypes, suggesting that the epitope may be a good target for vaccines. However, it is not clear whether the epitope for 1C19 is prM-dependent in common with other FLE-specific mAbs, as the epitope for 1C19 lies close to the FLE121.

E protein domain III. Antibodies targeting EDIII have been frequently isolated from mice71 and are among the most potent antibodies described to dengue virus, with 50% in vitro neutralization (NT50) levels at concentrations in the low picomolar range71,107,108,109,110,112,113,114,115. These antibodies are frequently specific for a single dengue virus serotype, although the EDIII-binding mAb 4E11 has been engineered to bind to and potently neutralize all four dengue virus serotypes122. The epitopes for some EDIII-binding mAbs have been mapped in detail by mutagenesis, crystallography and cryo-EM. The mAb E16 binds West Nile virus, a flavivirus related to dengue virus, where the epitope lies on the lateral ridge of EDIII71. Although not binding to dengue virus, E16 defines a class of antibodies binding to the lateral ridge of EDIII and shows potent neutralization of West Nile virus in vitro and in vivo in both prophylactic and therapeutic settings123. Another EDIII-specific antibody, 1A1D2, binds to the A strand (amino acids 305–312) of dengue virus EDIII124. In contrast to the lateral ridge of EDIII, the A strand is more conserved among the dengue viruses, allowing 1A1D2 to cross-react between DENV1, DENV2 and DENV3 (Ref. 111).

Interestingly, the epitope recognized by 1A1D2 is not fully exposed in the smooth mature form of the dengue virion, and cryo-EM has shown that a temperature-dependent conformational change allows EDIII to hinge up from its flat orientation in the smooth mature virion and facilitates mAb 1A1D2 binding124. Because of the potency of the EDIII-specific mAb and the general serotype specificity of the response, EDIII has been proposed as a potential immunogen in a variety of vaccine formulations. However, the contribution of EDIII to the human antibody response appears to be more limited; it has been shown that the depletion of EDIII-specific antibodies from human serum does not reduce its neutralization potential in vitro and in vivo77,125.

prM protein and virus maturity. As described above, immature dengue virus particles contain prM protein, which acts as a chaperone for E protein and prevents premature fusion to host cell membranes before virus release. prM protein is cleaved by furin protease in the Golgi and falls away when mature virus particles are released from cells4,7,10. However, prM cleavage is frequently incomplete, with a range of partially mature virus forms produced with intermediate levels of prM protein11. This produces a challenge for the host immune response, as mature and immature virus particles present markedly different structural determinants at the virion surface. Studies of primary human cells infected with dengue virus suggest that, following an insect bite, the human host is first infected by viruses that have a high prM content; however, viruses produced in human cells will have potentially lower prM content, although the prM content of dengue virus in ex vivo samples has not been measured15,16. Several dengue virus-specific antibodies, such as those targeting the FLE or prM protein, lack the ability to neutralize more mature, low-prM protein viruses15,16.

Alternative conformations of the virion surface. The binding of antibodies to viruses is a complex process, in which the packing of E protein and prM protein into the virion lattice affects the accessibility of the target epitopes to antibodies. Several antibodies that bind well to recombinant monomeric E protein fail to show good binding or neutralization of intact virions because the viral epitopes are poorly exposed14,17,71,124,126. The binding of such antibodies may be enhanced by the prM protein content of the virions, as described above for the FLE-specific antibodies, but may also be enhanced in vitro by prolonged incubation periods or increased temperature127. This has led to the proposition that the traditional view of the virion as an invariant or rigid, smooth or spiky entity (which is necessarily imposed by the solution of cryo-EM structures) is not accurate. Instead, it is proposed that the virion is a dynamic entity that is capable of adopting many different conformations with different thermodynamic stabilities, which is often referred to as virion breathing124,128,129. In this regard, an alternative 'bumpy' form of DENV2 particles has been recently described, in which there is disruption of the regular packing of the E protein dimers at the virion surface at 37 °C128,129. This has implications for antibody binding, as shown by cryo-EM structures of mAb 2D22 binding to DENV2 particles, in which the valency of binding to the bumpy form of the virus was lower than that to the more regular form130. The display of an 'ensemble' of different E protein and virion conformations can be continuously sampled and captured by antibodies, explaining the increase in binding of antibodies to dengue virus particles in a temperature- and time-dependent manner12,127.

Human mAbs. The study of antibody responses in humans has been revolutionized by the development of various techniques that have allowed the generation of large numbers of human mAbs. Three main techniques have been used: expansion and Epstein–Barr virus immortalization of memory B cells; single-cell sorting, cDNA cloning and antibody expression from plasmablasts isolated from acutely infected individuals; and optimization by electrofusion of traditional hybridoma technology to make antibodies from memory B cells131,132,133.

Antibodies to prM protein. prM protein-specific antibodies appear to be a major component of the memory B cell response to dengue virus infection; however, these antibodies show poor neutralization (maximum 30–50%) even at high concentrations15,134,135,136,137. prM protein-specific antibodies do not bind to fully mature virions as they lack prM protein, and many partially mature particles do not contain a high enough density of prM protein to allow neutralization; furthermore, they may be sufficient to promote ADE15,18. We have speculated that the inefficient cleavage of prM protein may be an immune evasion strategy, leading to the generation of poorly neutralizing antibodies directed to prM protein, as well as to the FLE, which also form a major component of the human antibody response15,16,118.

The high frequency, low potency and high ADE potential of antibodies directed to prM protein have implications for vaccine design. All attenuated vaccines at an advanced stage of development contain prM protein, whereas the ideal vaccine would focus responses to the E protein and minimize the prM protein component of the response to reduce the potential for ADE. One possible route to this would be the generation of attenuated chimeric viruses with prM protein derived from third party flaviviruses such as Japanese encephalitis or West Nile virus, in which the anti-prM protein response cross-reacts poorly with dengue virus15,138.

Antibodies targeting quaternary epitopes on the virion. Some of the most potent human antibodies bind to conformational and quaternary determinants that are only reproduced on intact dengue virions16,130,139,140,141. One well-characterized epitope, the so-called 'herring-bone epitope', has been described by cryo-EM for the DENV1-specific mAb HM14c10 (Ref. 139). This antibody binds to an epitope that bridges between two adjacent head-to-tail E protein dimers, which form a herring-bone-like conformation on the mature virion (Fig. 5a). Several other conformationally sensitive antibodies have been mapped by cryo-EM; the DENV1-specific mAb 1F4 binds to a single E protein monomer, but only in the context of the virion, whereas the DENV3-specific mAb 5J7 binds across three adjacent E protein monomers with a major component of the interaction across the hinge between EDI and EDII of the central E protein monomer140,143. These conformational antibodies show potent neutralization but are nevertheless serotype specific.

Figure 5: Quaternary epitopes.
figure 5

The cryo-electron microscopy (cryo-EM) structure shows where HM14c10 binds to the epitope between two envelope (E) protein monomers of adjacent dimers (part a). The structure (part b) and the cryo-EM reconstruction (part c) of the DENV2 particle in complex with an E protein dimer epitope (EDE)-specific monoclonal antibody (mAb) — 747(4) A11 and 747(4) B7 for part b and part c, respectively — demonstrate that the epitope of the antibody is across two E proteins within one dimer. The colour scale in part c indicates the radial depth from inside (red) to the outer shell (yellow, green and blue). Domains I, II and III of the E protein (EDI–EDIII) are indicated in red, yellow and blue, respectively. In part b top view, the green outline shows the footprint of the EDE-specific mAb. Figure in part a from Teoh, E. P. et al. The structural basis for serotype-specific neutralization of dengue virus by a human antibody. Sci. Transl. Med. 4, 139ra83 (2012). Reprinted with permission from AAAS. Figure in part b from Ref. 141, Nature Publishing Group. Figure in part c from Ref. 16, Nature Publishing Group.

PowerPoint slide

We have recently described the cloning of a large series of E protein-specific mAbs from patients infected with dengue virus16. One-third of the antibodies bind to a complex epitope present only on intact virions. Interestingly, although most of the panel of human antibodies showed good neutralization of high-prM protein viruses produced in insect cells, only binding of mAbs to the complex epitope on intact virions could fully neutralize low-prM protein virions produced in primary human cells.

These broadly neutralizing antibodies bind to the basic repeating E protein dimers that make up the virion surface lattice, and the epitope is termed the E protein dimer epitope (EDE)16,141 (Fig. 5b,c). The antibodies bind in a valley formed between the two monomers of E protein, making up the 90 head-to-tail dimers of the virion, and overlap with the footprint on E protein where prM protein sits on the immature dengue virus particle4. The antibodies make contact with several conserved amino acid side chains and main chain atoms in the E protein peptide backbone, which explains the cross-reactivity and broad neutralization of the four virus serotypes by these antibodies. Another recently reported mAb, 2D22, binds to E dimers, although the epitope is slightly shifted towards EDIII compared with the epitope described above, which results in 2D22 being specific for just DENV2 and not broadly cross-reactive130.

The discovery of the EDE opens up a number of interesting future possibilities for dengue treatment. Current vaccination strategies use tetravalent formulations with the aim of raising single serotype-specific responses against all four serotypes. The demonstration that potent and broadly neutralizing antibodies are produced during dengue virus infection indicates that the development of such antibodies should be a goal for the next-generation vaccines. Furthermore, it may be possible to design a universal, rather than tetravalent, formulation to achieve this response or to use heterologous prime–boost strategies. Importantly, as the response is limited to the E protein dimer, it opens the way for subunit vaccines that consist of E protein dimers alone. To achieve this, the E protein dimer will need to be stabilized, as E protein only forms dimers naturally at relatively high protein concentrations. A similar situation has been observed with respiratory syncytial virus, in which potent neutralizing antibodies bind to the trimeric pre-fusion forms of the fusion protein (F protein)144. Efforts to stabilize a soluble F protein trimer have been successful either by cavity-filling hydrophobic substitution or by covalent linkage of the F protein monomer, which allows for the generation of a novel subunit immunogen that is capable of inducing a neutralizing response against the F protein trimer in mouse and primate models of respiratory syncytial virus infection145.

As described here, the neutralization of dengue viruses is a complex process with several different virion forms produced during infection. Successful vaccines need to target potently neutralizing epitopes, such as those found on EDIII or quaternary epitopes, and minimize the targeting of poorly neutralizing epitopes, such as those on prM protein or the FLE.

Conclusion and future direction

Despite nearly 50 years of work, we still do not have a fully efficacious dengue virus vaccine, and it remains to be determined whether the Sanofi Pasteur CYD-TDV live attenuated vaccine will be licensed and deployed in endemic countries. However, the burden of dengue continues to increase and, despite promising advances in vector control strategies and an increasingly active search for antiviral drugs that target dengue virus, an effective vaccine is seen by many as the only realistic strategy to control the spread of this disease and reduce the burden it has on the health-care systems in endemic countries.

There are a number of second-generation live attenuated vaccines that are approaching larger-scale clinical trials. The challenge of inducing protection against all four virus serotypes using a tetravalent formulation is formidable, and heterologous prime–boost strategies, which mimic natural infection but have given conflicting results, may need to be evaluated further. The recent description of potent human neutralizing antibodies in dengue gives insight into the sort of responses that should be targeted by vaccines. However, major challenges in how to preferentially boost immune responses to these complex quaternary epitopes remain. Finally, the recent Sanofi Pasteur clinical trials have demonstrated the need for robust in vitro correlates of protection, which would guide the development of future vaccine trials, and there is now interest in developing dengue virus-challenge studies in humans to guide future vaccine development.